dr_frames

Pandas/DataFrame utilities for data manipulation, filtering, aggregation, and schema management.

Primitive quick reference for agents:

• primitives.columns: rename_columns, move_cols_to_beginning, strip_col_prefixes, drop_all_null_cols, drop_all_constant_cols, get_cols_by_prefix, get_cols_by_contains, move_cols_with_prefix_to_end, move_numeric_cols_to_end
• primitives.filtering: select_subset, filter_to_values, filter_to_range, make_filter_fxn
• primitives.ranking: select_best_by_metric
• primitives.coerce: coerce_numeric_cols, coerce_string_cols
• primitives.aggregation: aggregate_over_seeds, aggregate_by_group
• primitives.missing: fill_missing_values
• primitives.masked: masked_getter, masked_setter
• primitives.constant: get_constant_cols, get_groupwise_constant_cols
• primitives.unique: unique_non_null, unique_by_col, unique_by_cols
• primitives.namespaced: group_namespaced_values
• primitives.pipeline: maybe_pipe
• primitives.parsing: parse_list_string

Installation

uv add dr-frames

Quick Start

import pandas as pd
from dr_frames import (
    coerce_numeric_cols,
    filter_to_range,
    move_cols_to_beginning,
    select_subset,
)

df = pd.DataFrame({
    "name": ["alice", "bob", "charlie"],
    "value": ["1.0", "2.0", "3.0"],
    "category": ["x", "y", "x"],
})

result = (
    df.pipe(coerce_numeric_cols, ["value"])
    .pipe(select_subset, {"category": "x"})
    .pipe(filter_to_range, "value", 0.5, 2.5)
)

Flexible Schema

flexible_schema is the higher-level part of the library for working with dataframes whose columns you did not design yourself.

• DataField describes a logical field and can resolve which dataframe column it maps to.
• ComputedField describes a derived field that should be added from existing columns before downstream use.
• MetricDataField describes metric columns, especially prefixed metric names such as eval/....
• DataFormat is the container that ties those pieces together and gives you a reusable view of a dataframe schema.

Typical usage:

• build a DataFormat from a dataframe with DataFormat.from_df(...) or from a field-description mapping with DataFormat.from_dict(...)
• inspect unresolved fields and discovered metrics
• add computed fields for columns you want to derive once and reuse
• call prepare_for_plotting(...) to produce a dataframe restricted to the known fields, computed fields, and metrics
• use metric_col(...), get_metric(...), and get_config_columns(...) to drive plotting or grouped analysis code without hardcoding raw column names

Minimal example:

from dr_frames import ComputedField, DataField, DataFormat

fmt = DataFormat(
    fields=[
        DataField(id_string="model", column_name="model_name"),
        DataField(id_string="dataset"),
    ],
    computed_fields=[
        ComputedField(
            id_string="is_large",
            source_columns=["params_millions"],
            compute=lambda df: df["params_millions"] > 1000,
        )
    ],
)

plot_df = fmt.prepare_for_plotting(df)
config_cols = fmt.get_config_columns()

License

MIT